Roboflow Vision Evals — Detection mAP@75 (September)
Multimodal
Mean average precision at IoU 0.75, under the September observed protocol. One or three runs per effort tier; atomic sample count unreported.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 67.0 |
| Gemini 3.8 Flash | 62.5 |
| Gemini 3.7 Flash | 62.1 |
| Claude Opus 5.5 | 61.0 |
| Gemini 3.6 Flash | 58.1 |
| GPT-6 Sol | 57.8 |
| Gemini 3.5 Flash | 57.1 |
| Gemini 3.1 Pro Preview | 55.8 |
| Qwen3.8 Flash | 53.8 |
| Qwen3.7-Plus | 48.4 |
| GPT-6 Luna | 47.4 |
| Gemini 3.5 Flash-Lite | 46.6 |
| Qwen3.5-27B | 46.4 |
| GPT-5.6 Sol | 46.1 |
| GPT-5.6 Luna | 43.4 |
| Qwen3-VL-235B-A22B-Instruct | 42.9 |
| Muse Spark 1.2 | 42.1 |
| Kimi K3 | 41.5 |
| Muse Spark 1.1 | 40.9 |
| GPT-5.6 Terra | 40.6 |
| Claude Fable 5.1 | 40.4 |
| GLM 5V Turbo | 39.9 |
| Muse Spark 1.3 | 36.9 |
| MiMo-V2.6-Pro | 32.9 |
| Qwen3.7 Flash | 32.9 |
| MiMo-V2.6-Flash | 32.3 |
| Claude Fable 5 | 31.0 |
| Claude Opus 5 | 30.2 |
| Gemini 3 Flash Preview | 29.2 |
| Kimi K2.6 | 26.7 |
| Gemini 2.5 Pro | 25.1 |
| GPT-5.5 | 24.0 |
| Grok 4.7 | 23.7 |
| Muse Glimmer 30B | 22.3 |
| GLM-5.3 Flash | 17.9 |
| Sonnet 5 | 15.8 |
| Opus 4.8 | 13.3 |
| Grok 4.6 | 6.1 |
| Grok 4.5 | 5.0 |
| GPT-5.4 Mini | 3.3 |